server : use pytest-xdist for server tests (#28298)

* server : use pytest-xdist for server tests

This commit adds pytest-xdist to the server tests. This is pytest
plugin that distributes test execution across multiple CPU cores.

Assisted-by: pi:llama.cpp/qwen3.8-27B

Refs: https://github.com/ggml-org/llama.cpp/pull/26734#issuecomment-5220707042

* remove server_base_port and BASE_PORT

* use worksteal and pytest builting tmp_path

tools/server/tests/unit/test_compat_anthropic.py

@@ -21,7 +21,6 @@ def create_server():
     global server
     server = ServerPreset.tinyllama2()
     server.model_alias = "tinyllama-2-anthropic"
-    server.server_port = 8082
     server.n_slots = 1
     server.n_ctx = 8192
     server.n_batch = 2048
@@ -34,7 +33,6 @@ def vision_server():
     server = ServerPreset.tinygemma3()
     server.offline = False  # Allow downloading the model
     server.model_alias = "tinygemma3-anthropic"
-    server.server_port = 8083  # Different port to avoid conflicts
     server.n_slots = 1
     return server
 
@@ -1015,7 +1013,6 @@ def test_anthropic_thinking_with_reasoning_model(stream):
     server.jinja = True
     server.n_ctx = 8192
     server.n_predict = 1024
-    server.server_port = 8084
     server.start(timeout_seconds=600)  # large model needs time to download
 
     if stream:

tools/server/tests/unit/test_mcp_servers.py

@@ -37,7 +37,6 @@ def _start_server_with_mcp(mcp_json: str, **kwargs) -> ServerProcess:
     srv = ServerPreset.router()
     srv.server_tools = "all"
     srv.no_ui = True
-    srv.server_port = 8085  # avoid conflict with load_all() which uses 8080
     srv.mcp_servers_json = mcp_json
     for k, v in kwargs.items():
         setattr(srv, k, v)
@@ -183,7 +182,6 @@ def test_mcp_tools_not_listed_when_not_configured():
     server = ServerPreset.router()
     server.server_tools = "all"
     server.no_ui = True
-    server.server_port = 8085
     server.start()
 
     try:
@@ -250,7 +248,6 @@ def test_mcp_tools_via_json_config_file():
         server = ServerPreset.router()
         server.server_tools = "all"
         server.no_ui = True
-        server.server_port = 8085
         server.mcp_servers_config = config_path
         server.start()
 
@@ -468,7 +465,6 @@ def test_mcp_config_file_errors():
     server = ServerPreset.router()
     server.server_tools = "all"
     server.no_ui = True
-    server.server_port = 8085
     server.mcp_servers_json = "not valid json"
     try:
         server.start()
@@ -480,7 +476,6 @@ def test_mcp_config_file_errors():
     server = ServerPreset.router()
     server.server_tools = "all"
     server.no_ui = True
-    server.server_port = 8085
     server.mcp_servers_config = "/nonexistent/path.json"
     try:
         server.start()

tools/server/tests/unit/test_slot_save.py

@@ -10,10 +10,10 @@ STATE_FILE_HEADER_SIZE = 12
 server = ServerPreset.tinyllama2()
 
 @pytest.fixture(autouse=True)
-def create_server():
+def create_server(tmp_path):
     global server
     server = ServerPreset.tinyllama2()
-    server.slot_save_path = "./tmp"
+    server.slot_save_path = str(tmp_path)
     server.temperature = 0.0
 
 
@@ -94,7 +94,7 @@ def test_slot_restore_legacy_token_list():
     assert res.body["n_saved"] == 84
 
     # rewrite the token payload into a plain token list, as written by servers that predate the packed server_tokens format
-    path = os.path.join("tmp", "slot_legacy.bin")
+    path = os.path.join(server.slot_save_path, "slot_legacy.bin")
     with open(path, "rb") as f:
         data = bytearray(f.read())
 
@@ -462,7 +462,7 @@ def test_slot_save_restore_image_payload_larger_than_context(mmproj_server):
     })
     assert res.status_code == 200
 
-    path = os.path.join("tmp", "mm_slot_large_payload.bin")
+    path = os.path.join(server.slot_save_path, "mm_slot_large_payload.bin")
     with open(path, "rb") as f:
         data = bytearray(f.read())
     payload_size = struct.unpack_from("=I", data, STATE_FILE_HEADER_SIZE - 4)[0]

tools/server/tests/unit/test_tool_call.py

@@ -21,7 +21,6 @@ def create_server():
     global server
     server = ServerPreset.tinyllama2()
     server.model_alias = "tinyllama-2-tool-call"
-    server.server_port = 8081
     server.n_slots = 1
     server.n_ctx = 8192
     server.n_batch = 2048

tools/server/tests/unit/test_tools_builtin.py

@@ -64,11 +64,11 @@ def test_tools_builtin_read_file():
     assert "def test_tools_builtin_read_file" in text
 
 
-def test_tools_builtin_write_then_edit_file():
+def test_tools_builtin_write_then_edit_file(tmp_path):
     global server
     server.start()
 
-    log_path = os.path.join(PROJECT_ROOT, "test.log")
+    log_path = str(tmp_path / "test.log")
     try:
         write_res = call_tool("write_file", {"path": log_path, "content": "line1\nline2\nline3\n"})
         assert write_res["result"] == "file written successfully"
@@ -93,11 +93,11 @@ def test_tools_builtin_write_then_edit_file():
             os.remove(log_path)
 
 
-def test_tools_builtin_edit_file_rejects_non_unique_old_text():
+def test_tools_builtin_edit_file_rejects_non_unique_old_text(tmp_path):
     global server
     server.start()
 
-    log_path = os.path.join(PROJECT_ROOT, "test.log")
+    log_path = str(tmp_path / "test.log")
     try:
         call_tool("write_file", {"path": log_path, "content": "dup\ndup\n"})
         err = call_tool_expect_error("edit_file", {
@@ -275,11 +275,11 @@ def test_tools_builtin_docker_runtime_cleans_up_spawned_container():
     assert leftover.returncode != 0, f"container {container_id} was not cleaned up after server exit"
 
 
-def test_tools_builtin_edit_file_rejects_overlapping_edits():
+def test_tools_builtin_edit_file_rejects_overlapping_edits(tmp_path):
     global server
     server.start()
 
-    log_path = os.path.join(PROJECT_ROOT, "test.log")
+    log_path = str(tmp_path / "test.log")
     try:
         call_tool("write_file", {"path": log_path, "content": "line1\nline2\n"})
         err = call_tool_expect_error("edit_file", {